CN108351432A

CN108351432A - The focusing inverting of data-driven

Info

Publication number: CN108351432A
Application number: CN201580084177.7A
Authority: CN
Inventors: A·K·阮; F·A·马奥; T·维克; J·I·诺德斯卡格; E·考塞
Original assignee: Statoil Petroleum ASA
Current assignee: Equinor Energy AS
Priority date: 2015-08-27
Filing date: 2015-08-27
Publication date: 2018-07-31
Anticipated expiration: 2035-08-27
Also published as: EP3341767A1; EP3341767B1; AU2015406809A1; MX2018002495A; AU2015406809B2; CN108351432B; EP3341767A4; BR112018003716A2; CA2996840A1; WO2017034410A1

Abstract

The inverting of the focusing of data-driven discloses a kind of method for the geophysical data for handling and being acquired from subterranean zone.This method includes receiving the geophysical data acquired from subterranean zone, determine the data point high to quite sensitive degree, point executes inverting based on the data, wherein, during inverting, it is realized to abnormal focusing by increasing the weight of the data point, and determines at least one model that physical attribute is distributed in subterranean zone.

Description

Data-driven focus inversion

Technical Field

The present invention relates to geophysical exploration and data processing, and more particularly to a method for optimizing a physical property distribution model based on data acquired according to a geophysical exploration method (e.g., a controlled source electromagnetic or seismic survey).

Background

Geophysical data acquired from a subsurface region via conventional acquisition methods (e.g., seismic or CSEM or like surveys) may be used to obtain information about the distribution of physical properties in the subsurface region, so long as there is a mathematical relationship between the acquired data and the physical properties that are desired to be studied. Using the acquired data and a priori knowledge about the subsurface region, a model characterizing the distribution of the physical property may be generated. In this term, "model" refers to an estimated quantity. Inversion involves optimization of the generated model to determine the distribution of properties that explain the observed data.

The Inverse Problem (i.e. inverting Model parameters from the measured data set) is almost always an undetermined, non-unique and non-linear Problem (Tarantola, A. Inverse protocol Theory, Elsevier, New York 1987; Tarantola, A. Inverse protocol Theory and Methods for Model parameter estimation, Society for Industrial and Applied Mathematics 2005, and Zhdantv, M.S. geographic analysis Theory and Regulation schemes, Elsevier 2002). The measurement data does not provide enough information to ensure a unique solution at the desired resolution. Furthermore, the measurement data is also always affected by noise, inaccuracies and errors. Finally, the implementations of forward modeling operators required by inversion engines often have limited accuracy because they are based on assumptions for practical and computational reasons. To obtain a solution, artificial constraints and regularization on the model parameters are usually introduced to stabilize the inverse problem (Tarantola, 2005 and Zhdanov 2002). The assumptions behind these constraints and regularization are often inaccurate.

The resulting final inversion model is therefore a compromise between all these constraints, inaccuracies and errors. In practice, the inversion process always ends in an attraction domain with corresponding local minima, which favors certain parts and aspects of the inverse model while sacrificing other parts and aspects of the model. Which attraction domain and corresponding local minimum the inversion run actually ends with depends on the initial model, the regularization parameters, the included data points, the weights of the individual data points, and the optimization algorithm used. Without adding very strong assumptions/constraints, the results of this struggle between different inversion driving forces are difficult to control, potentially leading to severe artifacts in the inversion results if the assumptions are not accurate.

It is often desirable to reconstruct certain portions and aspects of the model as accurately as possible. The cost of a less accurate reconstruction of other less important parts/aspects of the model may then be accepted. Typically, the size of the survey object is very small compared to the entire model to be inverted. Correspondingly, the number of data points that are mainly sensitive to the large model background is completely dominant compared to several data points that are actually sensitive to the target. Thus, the target response tends to be too weak to compete with the other much stronger inversion driving forces, which results in poor target imaging and reasonably good background imaging.

As mentioned above, constraints and regularization are conventionally used to focus the inversion search space and results (Tarantola and ZHdanov2002), for example, to encourage the inversion engine to promote certain aspects of the inversion model, such as smoothness, density anomalies, low/high values for certain portions of the model (Tarantola and ZHdanov 2002). Hereafter, we use the term regularization to denote both constraint and regularization. This conventional approach is model-driven focused inversion (MDFI): which applies directly to the attributes of the model. The problem with this model-driven regularization is that the inverse model is forced to contain the properties required for regularization, whether or not those properties are present in the real model.

The model-driven focus inversion (MDFI) method can be shown as follows. Let the inversion/optimization problem find the M-dimensional model vector that minimizes the following mismatch function

Here, the data residualWherein,is a real or complex N-dimensional vector representing the measured data, andis used for modelForward modeling data. Sign of reverse vectorRepresenting a conjugate transpose. Data covariance matrixData errors and their dependencies are described. Conventionally, the pairs are formed by diagonal matrices containing data variancesAn approximation is made. Diagonal lineIndicating that all data points are irrelevant. Conventional used in geophysical inversionExamples of (ii) can be found in (Morten, j.p.,A.K.,(2009) CSEM data undirected analysis for3D inversion, SEG Expanded isolates 28,724).

Regularizing mismatch functionsOnly depending on the model parameters. λ is a real scalar used to control the strength of the model regularization term. Typically, the technician usesTo stabilize the inverse problem and control the inversion flow, i.e. to force/encourage the inverse model to have certain properties, such as smoothness, sharpness, parameters that remain within boundaries, proximity to a certain priority model, etc. (Tarantola,2005 and Zhdanov, 2002).

Disclosure of Invention

A data-driven focus inversion (DDFI) method is proposed that encourages the inversion optimizer to focus more on certain specific attributes, parameters, regions of the inversion model while sacrificing other attributes, parameters, and regions. Unlike MDFI, DDFI does not force the inversion model to contain the focused attributes if they are not present in the data.

Drawings

FIG. 1 shows the verticality (R) of the first and second forward models_V) And level (R)_H) The resistivity is abnormal.

FIG. 2 shows the electric field amplitude (E) of an azimuth receiver_xSolid line), phase (point), and noise estimate.

FIG. 3 shows the axial component E of a drag chain_xFocusing matrix with index n being 1/2 and f being 1HzIs provided.

fig. 4 shows a real model T, an initial model C, an α -0 inverted vertical resistivity model, and an α -20 inverted vertical resistivity model.

fig. 5 shows a real model T, an initial model C, an α -0 inverted horizontal resistivity model, and an α -20 inverted horizontal resistivity model.

fig. 6 shows an inverted vertical resistivity model inverted with α -0 and α -20 for the real model B, the initial model C.

fig. 7 shows an inverted vertical resistivity model inverted with α -0, α -5, and α -20 for the initial model C.

fig. 8 shows the vertical resistivity model inverted with α -0 and α -5 for the real model T, the initial model D.

fig. 9 shows the vertical resistivity model inverted with α -0 and α -20 for the real model T, the initial model E.

FIG. 10 shows a flow chart of an exemplary method.

FIG. 11 schematically illustrates, in block diagram form, an exemplary computer device.

FIG. 12 is a schematic diagram of a computer system for implementing one or more embodiments of the invention.

Detailed Description

The method according to the invention implements a geophysical data processing technique to generate a physical property distribution model of the subsurface region. Geophysical data acquired from a geophysical survey and a priori knowledge about the subsurface region are used to generate a model that characterizes the distribution of physical properties that produced the acquired data.

Some data points are highly sensitive to certain parameters, property regions/parts of the model. By increasing the weight of these data points during the inversion process, a more accurate physical property distribution model can be achieved, e.g., properties not present in the data do not become part of the inversion model.

Inversion involves adding new terms to the covariance matrix to encourage the inversion engine to focus its efforts on minimizing mismatch of selected data residuals that are highly sensitive to certain parameters, properties, regions/portions of the model. Hereafter, we will use the term object to denote a parameter, property or region/portion of the desired model we wish to study. We introduce a modified covariance matrix, which is defined as:

here, ,is a modified covariance matrix that is,is a matrix of units, and is,is the focus matrix and α is a constant used to control the focus intensity, in order to achieve focus on the target,should be defined as small/zero for data points that are not sensitive to the target and large for data points that are sensitive to the target. To achieve focus on the target, assume (given) is in the matrixThe most sensitive data points should have a higher weight (higher focus) than the data points that are not sensitive to the target.

In another embodiment of the invention, the focusing matrix may also be used in the reverse sense, i.e. willDefined as small for data points that are sensitive to certain parameters, attributes, regions of the inverse model and large at the remaining data points. This encourages the inversion engine to focus on non-target portions with minimal effort. This inverse approach can be used to reduce the influence of extraneous parts of the model in the inversion. Depending on the purpose of the inversion study, the "extraneous" part of the model may be, for example, salt, basalt, pipeline, etc.

For constructingThe method is toAnd passing the scattered field. Here two forward simulations are performed: once for background modelWith analog dataIs used for model at a timeWhich contains the same background model with the added target. The scattered field is then defined as:

therefore, the temperature of the molten metal is controlled,is a data response from the target only. Then, the user can use the device to perform the operation,can be defined, for example, as:

is a user-supplied matrix function that may be, for example,

where n is an index provided by the user.A matrix function of the scattered field provided for the user,for example:

wherein,to representAnd the matrix function diag () extracts the diagonal matrix.

It is to be noted that it is preferable that,the form ensures defocusing of the noise data points.

For diagonal lineAndthe effect of equation 7 is very similar to increasing the weight of the data points where the scattered field is strongest.

By way of example, a synthetic dataset may be used to show the effectiveness of the invention, and showing that data-driven focused inversion methods improve the inversion results around the target; the method is applicable to inversion of synthetic seafloor logging Controlled Source Electromagnetic (CSEM) datasets.

Electromagnetic data consists of electric and magnetic fields: are each [ E_x,E_y,E_z]And [ H_x,H_y,H_z]. The components of the electromagnetic field vector are complex in the frequency domain and depend on the source(s), receiver (r) and frequency (f), e.g. E_x→E_x(s,r,f)。

Conventionally, the data covariance matrix used in CSEM inversion is assumed to take advantage of the data uncertaintyThe elements given by the fixed factors are diagonal, examples may be found in (Morten, 2009;f.a. and Nguyen, a.k.enhanced surface response for marine CSEM rendering, Geophysics, No. 75, pages 7-10, 2010). Here we define the variance for different electromagnetic fields as:

wherein, beta₁is the autocorrelation multiplication uncertainty between the field components, and β₂and beta₃Maao and Nguyen (2010) provide the reasons and explanations behind multiplicative uncertainty and additive noiseUp to 6 additional components are provided, and then the corresponding diagonal covariance matrix for the electric field portion is:

similarly, for the magnetic field:

synthetic data examples

In this section we will show that the data-driven focus inversion method is well suited for3D anisotropic inversion on synthetic seafloor logging controlled source electromagnetic data sets.

The synthetic model and synthetic dataset are largely generated from measured data from a real survey. The synthetic model is created using the true seismic horizons at the survey location. Seismic-driven CSEM inversion from real datasets (Nguyen, a.k.,hansen J.O, (2013) CSEM expansion in the Barents Sea, Part II-high resolution CSEM inversion,75th Annual Conference and inhibition, EAGE, ExtendedAbstracts, We0403) results. Error! (reference source not found) shows slices of the real model (T) and the background model (B). The model T has a vertical resistivity R therein_VAn anomaly; while no anomalies exist in model B. This R in the real model_VAnomalies are the target of inversion studies. Horizontal resistivity model R_HThe same for both model T and model B. R_HThe reason behind the lack of anomalies in (A) is due to the fact that thin horizontal resistive anomalies have virtually no effect on the CSEM data, so R is not constructed in the seismic-driven CSEM inversion_HAnd (6) abnormal. We believe that it is not necessary to artificially associate R with_HAddition of anomalies to a real modelIn (1). 3D forward modeling with grid dx-dy-100 m and dz-40 m is based on the finite difference time domain method (F.A. (2007) Fast fine-difference time-domain modifying for marine-subsurface electronic schemes, geophisics, 72, A19-23). Data were collected for frequencies f of 0.25Hz, 0.5Hz, 1 Hz. The water depth measurements are taken into account in the modelling.

The navigation and orientation of the transmitter and receiver is taken from the survey data. For each synthetic data point, we add to it a complex number with a random phase and an amplitude given by the noise estimate for the corresponding real data point. Furthermore, we distort the receiver orientation values by uniformly distributing random numbers between-3 and 3 degrees. Thus, noise levels, receiver rotation errors, navigation defects are all close to the uncertainty and condition of the true acquisition. The electric field of the azimuth line affected by noise is wrong! (reference source not found). We see that additive noise can significantly affect the data in both amplitude and phase, particularly where offsets of the magnetotelluric pulse train occur.

In order to construct a covariance matrix of the focus of the object, the sensitivity of each data point to the object needs to be estimated. This can be achieved by comparing clean synthetic data from model T and model B.

Error! (reference source not found) shows the axial component E of a drag chain_xFocusing matrix with index n being 1/2 and f being 1HzIs provided. We see that the target clearly projects its presence and information into a limited set of data points.

Inversion

The L-BFGS-B regimen (Zach, J.J.,A.K.,t, and3D inversion of marine CSEM data using a fast fine-difference time-domain for forward coding and improved knowledge based optimization SEG Expanded abstract 27,614,2008) was used to invert the synthetic datasets from model T and model B. No regularization is applied in this pixel-based focus inversion. The initial model C is a smooth and enlarged version of model B with dx-dy-150 m and dz-40 m. For inversion, we use field E_xAnd E_yAnd the frequencies f are 0.25Hz, 0.5Hz, 1Hz, and the corresponding maximum absolute offsets are 10000m, 9000m, 8000 m. The minimum offset is 1500m for all frequencies. The final inversion result is obtained from iteration 100, unless explicitly mentioned.

the error! (reference source not found) shows the true, initial model C, the vertical resistivity model for α ═ 0 inversion and the vertical resistivity model for α ═ 20 inversion we see that the α ═ 0 inverted model contains anomalies at the correct locations.

the error | (reference source not found) shows the true, initial model C, the horizontal resistivity model inverted at α ═ 0, and the horizontal resistivity model inverted at α ═ 20 we see that the inverted models are more or less the same and very similar to the initial model.

the question that is naturally asked is whether the weight of the focal anomaly is likely to introduce an artificial anomaly in the inversion results even when there is no such anomaly in the input data fig. 6 shows an inverted model from dataset B for α 0 and α 20.

fig. 7 shows the inverted vertical resistivity model for α -0, α -5 and α -20 it can be seen that increasing the focus also increases the clarity of the resistivity anomalies in the inverted model.

Another natural problem is how can a significantly wrong initial model affect the inversion results when applying weights for focus anomalies? We investigated this problem by performing an inversion with initial models D and E, which are distorted versions of the "correct" initial model C. The initial model D is R_V→R_VX 0.8 and R_H→R_HModified version of model C by 0.8.

the error | (reference source not found) shows the vertical resistivity model for the case of true, original model D, inverted with α -0, and inverted with α -5 we see that when focused weights are used in the inversion, the resistivity anomaly is restored again much better than conventional unfocused weights.

Error! (reference source not found) 9 shows a reference tovertical resistivity models for cases where true, initial model E, inverted with α ═ 0, and inverted with α ═ 5_v→R_vX 1.2 and R_H→R_HModified version of model C by 1.2. We see that when focused weights are used in the inversion, the resistivity anomaly is again recovered much better than the conventional unfocused weights. The inverted horizontal resistivity models corresponding to the model shown in fig. 8 are more or less identical.

Thus, it can be concluded that: a significantly wrong initial model does not reduce the effect of the weighting of the focus anomalies.

Above we have shown that the data-driven focus inversion (DDFI) method is well suited for CSEM inversion.

Unlike conventional model-driven focus inversion (MDFI) methods, DDFI does not impose certain properties into the inverted model if the presence of these properties is not present in the data.

Inversion of the measurement data often provides results with a lower resolution than required for interpretation, or the estimated parameters have a higher uncertainty than required, especially for a small target area or a small set of important parameters. Data-driven focused inversion places more effort on reducing the data residuals of selected data points that are sensitive to user-defined targets. Recall that the term object herein refers to selected attributes, parameters and regions of the model. The inversion results will therefore provide better clarity of the target at the expense of a generally acceptably poor background from the user's perspective.

The inversion almost always ends with a local minimum. Which minimum it ends with depends on the initial model, regularization, optimizer algorithm. Model-driven focus inversion enforces the properties indicated by the regularization term into the inversion model, e.g., a smooth regularization term forces the inversion model to be smooth regardless of the measured data. In contrast, data-driven focus inversion only puts more effort on reducing the data residuals for selected data points. Thus, if no target response is found in the measurement data, the focus function for the target does not force the inverse model to contain the target. Conversely, DDFI will ensure that the inverse model does not contain targets in this case, since the weighted data mismatch will be too high. Therefore, DDFI increases the probability that the inversion ends with a local minimum, which ensures that the properties of the target are accurately described.

Data-driven focus inversion DDFI attempts to find a model that minimizes the data fit for selected data points that are sensitive to a target, at the expense of other data points that are only sensitive to background and not to the target. This enables a more accurate definition of the target.

Fig. 10 is a flowchart illustrating exemplary steps according to the first embodiment. The following numbering corresponds to that of fig. 10:

s1, receiving geophysical data collected from a subsurface region.

And S2, determining data points with high abnormal sensitivity.

S3, performing inversion based on the data points, wherein during the inversion, focusing on the anomaly is achieved by increasing the weight of the data points.

S4, determining at least one model of physical attribute distribution in the underground region.

FIG. 11 is a flowchart showing exemplary steps according to a second embodiment; the following numbering corresponds to that of fig. 11:

s5, receiving the geophysical data collected from the underground area.

And S6, determining data points with low abnormal sensitivity.

S7, performing inversion based on the data points, wherein during the inversion, focusing on the anomaly is achieved by reducing the weight of the data points.

S8, determining at least one model of physical attribute distribution in the underground region.

Referring to FIG. 12, an exemplary computer device 1 capable of performing a method of processing geophysical data acquired from a subsurface region is provided, as described in an embodiment of the present invention. The computer device may contain or be programmed by a computer program 2, the computer program 2 comprising computer readable instructions for processing geophysical data acquired from a subsurface region, as described in embodiments of the invention. The computer program 2 may be stored on a non-transitory computer readable medium in the form of a memory 3. The computer program 2 may be provided from an external source 4. The computer device 1 may comprise a display 5 allowing a user to visualize the model of the distribution of the physical property in the determined subsurface region.

Claims

1. A method of processing geophysical data acquired from a subsurface region, the method comprising:

receiving geophysical data acquired from a subsurface region;

determining data points having a high sensitivity to anomalies;

performing an inversion based on the data points, wherein during the inversion focusing on the anomaly is achieved by increasing a weight of the data points, an

At least one model of a distribution of physical properties in the subsurface region is determined.

2. The method of claim 1, wherein the geophysical data may be any one of electromagnetic data, controlled source electromagnetic data, seismic data, gravity data, or resistivity data.

3. The method of claim 1 or 2, wherein increasing the weight of a data point is achieved by defining a covariance matrix of the form:

wherein,is a focus matrix and α is a constant for controlling the focus intensity and is added for the data points

4. The method of any one of claims 1, 2 or3, wherein a high sensitivity of a data point to an anomaly is determined by comparing two forward models, which are identical except for the anomaly.

5. The method of any preceding claim 4, wherein the anomaly may be any one of a parameter, property or region/portion of a model.

6. A method of processing geophysical data acquired from a subsurface region, the method comprising:

receiving geophysical data acquired from the subsurface region;

determining data points having low sensitivity to anomalies;

performing an inversion based on the data points, wherein during the inversion, focusing on the anomaly is achieved by reducing a weight of the data points; and

7. The method of claim 6, wherein the geophysical data may be any one of electromagnetic data, controlled source electromagnetic data, seismic data, gravity data, or resistivity data.

8. The method of claim 6 or 7, wherein increasing the weight of a data point is achieved by defining a covariance matrix of the form:

9. The method of any of claims 6, 7 or 8, wherein a low sensitivity of data points to anomalies is determined by comparing two forward models, the two forward models being the same except for the anomalies.

10. The method according to any one of claims 6 to 9, wherein the anomaly may be any one of a parameter, property or region/portion of a model.

11. A computer program arranged to control a computer to perform a process in a method according to any one of claims 1 to 10.

12. A computer containing a program according to claim 11 or programmed by a program according to claim 11.

13. A computer readable medium containing a program according to claim 12.

14. Transmission of a program according to claim 11 over a network.